Merge write benchmarks

[vweevers]

Changes:

• Merges write, write-sorted, write-random into a single put benchmark (closes Merge write, -sorted and -random benchmarks into one #6, closes Don't use monotonic-timestamp for write-sorted benchmark #9), using options like --keys seq to switch between random and sequential keys (as well as strings and buffers)
• No longer uses crypto (less blocking)
• Records average time per 1000 ops (like write, but unlike write-sorted and write-random which did not aggregate)
• Uses Buffer.byteLength() to compute total bytes written (closes Reported bytes written is incorrect (too low) #8)
• Increases accuracy by using x.toFixed(3) instead of Math.floor(x)
• I wanted to make the throughput metric an average of a certain window but it'd require a separate data stream which complicates the gnuplot integration too much, so I opted to document the "type of average" instead (closes Maybe use moving average for mb/s #10 for now).

Example of what this enables: compare write perf of sequential vs random keys:

Click to expand

level-bench run put leveldown
level-bench run put leveldown -b [--keys seq]
level-bench run put leveldown -b [--keys seqReverse]
level-bench plot put

[vweevers]

I'm gonna go ahead and merge this, because there are bigger fish to fry (for which we actually need these benchmarks) and this project is in its infancy anyway. My use of terms like "Simple Moving Average" may be incorrect (IANAMathematician), that can be fixed later if need be.

@peakji @ralphtheninja LMK if you disagree and want me to slow down to give you time to review.